CN108074232A - A kind of airborne LIDAR based on volume elements segmentation builds object detecting method - Google Patents
A kind of airborne LIDAR based on volume elements segmentation builds object detecting method Download PDFInfo
- Publication number
- CN108074232A CN108074232A CN201711362019.0A CN201711362019A CN108074232A CN 108074232 A CN108074232 A CN 108074232A CN 201711362019 A CN201711362019 A CN 201711362019A CN 108074232 A CN108074232 A CN 108074232A
- Authority
- CN
- China
- Prior art keywords
- volume elements
- airborne lidar
- building
- voxel
- laser point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
The present invention proposes that a kind of airborne LIDAR based on volume elements segmentation builds object detecting method, and this method is:Original airborne LIDAR cloud data is read, forms original airborne LIDAR cloud data collection;Original airborne LIDAR cloud data rule is turned into gray scale 3D voxel data collection;Based on connective and radiation characteristic similarity criterion, gray scale 3D voxel datas are split and are labeled as several 3D connected regions;Characteristic based on building roof and facade detects the 3D connected regions that building roof and facade are formed successively, completes the airborne LIDAR based on volume elements segmentation and builds analyte detection;This method make use of the neighborhood relationships implied in 3D voxel datas between each volume elements and the characteristic of building well, contribute to the airborne LIDAR Point Cloud Processing based on the solid unit modeling theory and the development of application.
Description
Technical field
The invention belongs to Remote Sensing Data Processing technical fields, and in particular to a kind of airborne LIDAR based on volume elements segmentation is built
Build object detecting method.
Background technology
Building is component indispensable in 3D geo-information products, thus automatic, high-precision and is quickly built
Building object target detection becomes research hotspot.Airborne laser radar (Light Detection And Ranging, LIDAR) skill
Art is capable of providing very three-dimensional (3Dimension, 3D) cloud data intensive, accurate, with Geographic Reference, and includes
The strength information of echo-signal.Thus target detection of the airborne LIDAR data particularly suitable for 3D.Classical building analyte detection
Method can be divided into according to its data structure used:Building quality testing based on discrete point cloud, grid grid and irregular triangle network
Survey method.The primitive character of airborne LIDAR data has been fully retained in point cloud structure, is true 3D data.But each laser in the structure
Point is relatively independent, does not explicitly record the adjacency information on each laser point, it is impossible to directly obtain data processing needs
Neighborhood information, thus causes data processing algorithm difficult design, and operational efficiency is relatively low.In addition, the building analyte detection based on cloud
As a result it is not easy to realize vector quantization.The same plane of grid grid and irregular triangle network (X, Y) coordinate can only correspond to an elevation
(Z) value, the expression of such data structure certainly exists information loss for 3D LIDAR cloud datas, and then influences based on the knot
The integrality of the object detection results of structure.In addition, the testing result of the building object detecting method based on grid grid is 2D forms.
As it can be seen that data structure is unfavorable for playing the technical advantage of the true 3D of airborne LIDAR used by classical building object detecting method.Body
Metadata structure is a kind of true 3D data structures, and information loss will not be caused by expressing LIDAR cloud datas with it.The structure simultaneously
Geometric topo-relationship is implied between internal volume elements, thus the design of the data processing algorithm based on the data structure is relatively easy.
Forestry or the detection of ground target are more common in the analysis of airborne LIDAR data based on volume elements structure, it is of the invention then innovatively
Volume elements structure is combined with building target detection, it is proposed that the 3D building target detection methods based on volume elements.
The content of the invention
In view of the deficiencies of the prior art, the present invention proposes that a kind of airborne LIDAR based on volume elements segmentation builds analyte detection side
Method.
A kind of airborne LIDAR based on volume elements segmentation builds object detecting method, comprises the following steps:
Step 1:Original airborne LIDAR cloud data is read, forms original airborne LIDAR cloud data collection;
Step 2:Original airborne LIDAR cloud data rule is turned into gray scale 3D voxel data collection;
Step 2.1:The rejecting abnormalities data from original airborne LIDAR cloud data obtain removal abnormal data set;
Step 2.2:Removal abnormal data set rule is turned into gray scale 3D voxel data collection;
Step 3:Based on connective and radiation characteristic similarity criterion, gray scale 3D voxel datas are split and are labeled as several
A 3D connected regions;
Step 4:Characteristic based on building roof and facade detects building roof and is connected with the 3D that facade is formed successively
Region completes the airborne LIDAR based on volume elements segmentation and builds analyte detection;
Step 4.1:Based on area performance, elevation hopping behavior and density feature, the 3D connected regions of building roof are detected
Building roof detection is completed in domain;
Step 4.2:Based on buffer zone analysis, the 3D connected regions of building facade are detected, complete the detection of building facade.
2nd, the airborne LIDAR according to claim 1 based on volume elements segmentation builds object detecting method, and feature exists
In the step 2.1 specifically comprises the following steps:
Step 2.1.1:The frequency of each laser point height value in original airborne LIDAR cloud data is counted, and with Nogata
The form visualization of figure shows statistical result;
Step 2.1.2:Determine highest elevation threshold value T corresponding with real terrainhWith lowest elevation threshold value Tl;
Step 2.1.3:For each laser point in original airborne LIDAR cloud data, if its height value is high higher than highest
Journey threshold value ThOr less than lowest elevation threshold value Tl, then the laser point is abnormal data, is rejected, otherwise retains the laser point, most
Removal abnormal data set is obtained eventually.
The step 2.2 specifically comprises the following steps:
Step 2.2.1:Three dimensions scope is represented with the oriented bounding box of removal abnormal data set;
Step 2.2.2:According to removal abnormal data the equalization point spacing of laser point is concentrated to determine volume elements in the x, y, z-directions
Resolution ratio (Δ x, Δ y, Δ z), i.e. voxel size;
Step 2.2.3:According to voxel resolution, (Δ x, Δ y, Δ z) divide oriented bounding box, obtain 3D volume elements
Grid, each 3D volume elements grid unit is volume elements;
Step 2.2.4:Each laser point is concentrated to be mapped to 3D volume elements grid removal abnormal data, and then according to 3D volume elements
The Intensity attribute of the laser point included in grid is each volume elements assignment, obtains gray scale 3D voxel data collection.
The step 3 comprises the following steps:
Step 3.1:The non-zero value volume elements that gray scale 3D voxel datas are concentrated is scanned successively, is not labeled until k-th non-zero
It is worth volume elements, wherein, k=1,2 ...;
Step 3.2:Using depth-first strategy traversal connected with k-th of non-zero value volume elements 3D and voxel values similar in it is all
Volume elements, and labeled as Ll, wherein, l is the index of mark label, wherein, l=1,2 ...;
Step 3.3:The not labeled non-zero value volume elements of gray scale 3D voxel datas concentration is continued to scan on, until all bodies
Member is all labeled, obtains several 3D connected regions;
The difference of volume elements, that is, voxel values similar in the voxel values is less than two volume elements of gray difference threshold.
The step 4.1 specifically comprises the following steps:
Step 4.1.1:The 3D connected regions of non-building roof are rejected based on area performance;
Step 4.1.2:The 3D connected regions of non-building roof are rejected based on elevation hopping behavior;
Step 4.1.3:The 3D connected regions of non-building roof are rejected based on density feature.
The step 4.2 specifically comprises the following steps:
Step 4.2.1:Detect the profile of each building roof;
Step 4.2.2:In the horizontal plane, centered on any building roof profile, with each individual in inner side and outer side
Member establishes buffering area for width;
Step 4.2.3:To any 3D connected regions, if it is located inside buffering area and its gray value is corresponding with the region
The difference of the gray value of building roof is less than gray difference threshold, then is determined as building facade.
The Intensity attribute of the laser point included in the volume elements grid according to 3D is as follows for the detailed process of each volume elements assignment
It is shown:
Volume elements containing laser point is assigned a value of laser point strength mean value, the volume elements for not containing laser point is assigned a value of 0, into
Each volume elements assignment discretization to { 0 ..., 255 }, is obtained each voxel values by one step.
Beneficial effects of the present invention:
The present invention proposes that a kind of airborne LIDAR based on volume elements segmentation builds object detecting method, and this method first will be airborne
LIDAR cloud data rules turn to voxel data, then divide voxel data according to connective and radiation characteristic similarity criterion
It cuts and is labeled as several 3D connected regions, finally detected successively by building room using the characteristic of building roof and facade
The 3D connected regions that top, facade are formed.Volume elements builds quality testing compared with traditional point cloud, grid grid and irregular triangle network etc.
Common discrete LIDAR cloud datas expression way in survey is a kind of simple apparent space structure expression and is based on the table
The data processing algorithm design reached is easier.Building object detecting method proposed by the present invention based on volume elements segmentation, is by point
Point cloud segmentation and target detection in cloud data are converted into the target detection based on volume elements segmentation, make use of 3D volume elements numbers well
The neighborhood relationships and the characteristic of building implied between each volume elements, contribute to the airborne LIDAR point based on the solid unit modeling theory
Cloud data processing and the development of application.
Description of the drawings
Fig. 1 is the flow of the airborne LIDAR building object detecting method based on volume elements segmentation in the specific embodiment of the invention
Figure;
Fig. 2 is original airborne LIDAR cloud data in the specific embodiment of the invention;
Wherein, (a) is Area2 cloud datas, and (b) is Area3 cloud datas, and (c) is the corresponding figure of Area2 cloud datas
Picture, (d) are the corresponding image of Area3 cloud datas;
Fig. 3 is that original airborne LIDAR cloud data rule is turned to gray scale 3D volume elements numbers in the specific embodiment of the invention
According to the flow chart of collection;
Fig. 4 is gray scale 3D voxel data top views obtained by the regularization of LIDAR cloud datas in the specific embodiment of the invention;
Wherein, (a) is the corresponding voxel datas of Area2, and (b) is the corresponding voxel datas of Area3;
Fig. 5 is to split gray scale 3D voxel datas in the specific embodiment of the invention and be labeled as several 3D connected regions
Flow chart;
Fig. 6 is LABEL (k, u, L in the specific embodiment of the inventionl) program process;
Fig. 7 is gray scale 3D voxel data segmentation result top views in the specific embodiment of the invention;
Wherein, (a) is the segmentation result of Area2, and (b) is the segmentation result of Area3;
Fig. 8 is the gray scale frequency histogram of Area3 in the specific embodiment of the invention;
Fig. 9 is that the flow that the airborne LIDAR based on volume elements segmentation builds analyte detection is completed in the specific embodiment of the invention
Figure;
Figure 10 is the schematic diagram of setting buffers in the specific embodiment of the invention;
Wherein, dark volume member represents the building roof profile of extraction, and Dark grey volume elements represents inside buffering area, light grey
Volume elements represents outside buffering area;
Figure 11 is that the airborne LIDAR building testing result based on volume elements segmentation is completed in the specific embodiment of the invention;
Wherein, (a) is the testing result of Area2, and (b) is the testing result of Area3.
Specific embodiment
The specific embodiment of the invention is described in detail below in conjunction with the accompanying drawings.
In present embodiment, used on CPU dual-core 2.4GHz, 7 flagship edition system of memory 4GB, Windows
This method is realized in the programming of MATLAB 7.11.0 platforms, and further passes through the effective of the accuracy assessment verification method to this method
Property.
A kind of airborne LIDAR based on volume elements segmentation builds object detecting method, as shown in Figure 1, comprising the following steps:
Step 1:Original airborne LIDAR cloud data is read, forms original airborne LIDAR cloud data collection.
In present embodiment, using International Photography measurement and remote sensing association (International Society for
Photogrammetry and Remote Sensing, ISPRS) working groups of Section III/4 provide two groups (Area2 and Area3,
As shown in Figure 2) dedicated for target classification test of heuristics city sample data as experimental data, with the effective of the method for inspection
Property and feasibility.Experimental data obtains (500 meters of flying height, field angle 45) by LeicaALS50 airborne lidars instrument.This two groups
Residential block in data comprising the high-rise urban residential building object surrounded by trees and with small annex.Cloud data is close
It spends for 4 points/m2。
In present embodiment, original airborne LIDAR cloud data collection P={ p are definedi(xi, yi, zi), i=1 ..., n },
Wherein, i is the index of original airborne LIDAR cloud data, and n is the number of original airborne LIDAR cloud data, piIt is i-th
Original airborne LIDAR cloud data, coordinate are (xi, yi, zi)。
Step 2:Original airborne LIDAR cloud data rule is turned into gray scale 3D voxel data collection, idiographic flow such as Fig. 3
It is shown.
Step 2.1:The rejecting abnormalities data from original airborne LIDAR cloud data obtain removal abnormal data set.
Step 2.1.1:The frequency of each laser point height value in original airborne LIDAR cloud data is counted, and with Nogata
The form visualization of figure shows statistical result.
Step 2.1.2:Determine highest elevation threshold value T corresponding with real terrainhWith lowest elevation threshold value Tl。
Step 2.1.3:For each laser point in original airborne LIDAR cloud data, if its height value is high higher than highest
Journey threshold value ThOr less than lowest elevation threshold value Tl, then the laser point is abnormal data, is rejected, otherwise retains the laser point, most
Removal abnormal data set is obtained eventually.
In present embodiment, removal abnormal data set is denoted as Q={ qi′(xi′, yi′, zi′), i '=1 ..., t }, wherein, i '
It is to remove the index that abnormal data concentrates data, t is to remove the number that abnormal data concentrates data, qi′It is removal abnormal data set
In the i-th ' a data, coordinate be (xi′, yi′, zi′)。
In present embodiment, highest elevation threshold value ThWith lowest elevation threshold value TlFor constant, value need to be according to original airborne
The space distribution situation of LIDAR cloud datas determines.
Step 2.2:Removal abnormal data set rule is turned into gray scale 3D voxel data collection.
Step 2.2.1:Three dimensions scope is represented with the oriented bounding box of removal abnormal data set.
In present embodiment, the oriented bounding box of removal abnormal data set Q is cuboid, and the big I in bottom surface is by asking for
The minimum enclosed rectangle that removal abnormal data set Q is projected on X/Y plane determines that height can be by (zmax-zmin) determine.
Wherein, zmaxIt is the maximum for removing the z coordinate of laser point in abnormal data set Q, zminIt is removal abnormal data set Q
The minimum value of the z coordinate of middle laser point, zmax=max { zi′, i '=1 ..., t }, zmin=min { zi′, i '=1 ..., t }.
Step 2.2.2:Determine volume elements in x, y, z direction according to the equalization point spacing of laser point in removal abnormal data set Q
On resolution ratio (Δ x, Δ y, Δ z), i.e. voxel size.
In present embodiment, volume elements resolution ax x in the x, y, z-directions, Δ y, calculation formula such as formula (1) institute of Δ z
Show:
Wherein, Sxy={ (xi′, yi′), i '=1 ..., t } it is obtained by projections of the removal abnormal data set Q on XOY plane
Two-dimentional point set, C (Sxy) it is point set SxyConvex hull, A (C (Sxy)) it is convex hull C (Sxy) area.
Step 2.2.3:According to voxel resolution, (Δ xx, Δ y, Δ z) divide oriented bounding box, obtain 3D volume elements
Grid, each 3D volume elements grid unit is volume elements.
In present embodiment, based on voxel resolution, (oriented bounding box can be divided into 3D bodies by Δ xx, Δ y, Δ z)
First grid is represented with 3D volume elements arrays.If V is the volume elements set in 3D volume elements arrays, as shown in formula (2):
V={ vj(rj, cj, lj), j=1 ..., m }, (2)
Wherein, j is volume elements index;M is volume elements number;vjIt is the voxel values of j-th of volume elements;(rj, cj, lj) it is j-th of volume elements
Coordinate (row, column and level number) in volume elements array.Volume elements number in X-direction is R, and the volume elements number in Y-direction is C, Z
Volume elements number on direction is L.Wherein, R, C, L are by formula (3) Suo Shi:
Wherein,For the operator that rounds up, xmax=max { xi′, i '=1 ..., t }, xmin=min { xi′, i '=
1 ..., t }, ymax=max { yi′, i '=1 ..., t } and .ymin=min { yi′, i '=1 ..., t }.
It therefore deduces that, shown in volume elements number m such as formulas (4):
M=R*C*L (4)
Step 2.2.4:Each laser point in removal abnormal data set Q is mapped to 3D volume elements grid, and then according to 3D bodies
The Intensity attribute of the laser point included in first grid is each volume elements assignment, obtains gray scale 3D voxel data collection.
In present embodiment, each laser point in removal abnormal data set Q is mapped to 3D volume elements grid, and then according to 3D
The Intensity attribute of the laser point included in volume elements grid is each volume elements assignment.Wherein, the volume elements containing laser point is assigned a value of swashing
The volume elements for not containing laser point is assigned a value of 0 by spot intensity average, as shown in formula (5):
Wherein,It is accorded with for downward floor operation.Further by each volume elements assignment discretization to { 0 ..., 255 }, obtain each
Voxel values.Gray scale 3D voxel data collection is obtained as a result, completes the regularization to removing abnormal data set.
In present embodiment, the voxel data top view obtained by cloud regularization as shown in figure 4, gray value represent it is different
Volume elements intensity value, it can be seen that the intensity value of building roof is relatively uniform and is had differences with the intensity value of other targets.
Step 3:Based on connective and radiation characteristic similarity criterion, gray scale 3D voxel datas are split and are labeled as several
A 3D connected regions, idiographic flow are as shown in Figure 5.
Step 3.1:The non-zero value volume elements that gray scale 3D voxel datas are concentrated is scanned successively, is not labeled until k-th non-zero
It is worth volume elements, wherein, k=1,2 ....
Step 3.2:Using depth-first strategy traversal connected with k-th of non-zero value volume elements 3D and voxel values similar in it is all
Volume elements, and labeled as Ll, wherein, l be mark label index, l=1,2 ....
It is v for voxel values in present embodimentkK-th of volume elements of=u, using depth-first strategy traversal and k-th
All volume elements similar in non-zero value volume elements 3D connections and voxel values, as shown in fig. 6, caller LABEL (k, u, Ll) mark with
The label of volume elements is L similar in kth non-zero value volume elements 3D connections and intensity valuel.Volume elements, that is, voxel values similar in voxel values it
Difference is less than two volume elements of intensity difference threshold value Tg.
Step 3.3:The not labeled non-zero value volume elements of gray scale 3D voxel datas concentration is continued to scan on, until all bodies
Member is all labeled, obtains several 3D connected regions.
In present embodiment, the target of segmentation is exactly to connect the cell combination connected and radiation characteristic is similar to a 3D
Region.Assuming that 3D volume elements arrays V is altogether comprising l 3D connected region, the task of segmentation be exactly l label is distributed to it is each in V
A volume elements to belong to the volume elements of same 3D connected regions with identical label, and belongs to different 3D connected regions
Volume elements then has different labels.It is same category of think of based on volume elements similar in connection and intensity value in specific implementation process
Think, split V using 3D connected components labeling algorithm and be labeled as l 3D connected region, obtained segmentation result such as Fig. 7 institutes
Show.It can be seen that most volume elements for belonging to single building are all divided in a 3D connected region.
In present embodiment, different spatial neighborhood scales and different intensity difference threshold values are applied in above-mentioned labeling process
It can obtain different segmentation results simultaneously and then influence follow-up building extraction result.Optimal spatial neighborhood scale will in an experiment
It determines.Suitable strength difference threshold value is determined (by taking Area3 in experimental data as an example) by following proposal:
The frequency of the gray value of the non-zero value volume elements in 3D volume elements arrays V is counted, and is shown with represented as histograms, such as Fig. 8 institutes
Show, 1,28,81 and 133 are corresponded to comprising 4 normal distributions, peak Distribution in figure.In order to ensure the volume elements for belonging to single building
A 3D connected region is divided to, the 3rd normal distribution corresponding to building is used to calculate optimal difference limen of intensity
Value Tg.The scope of 3rd normal distribution gray value is [61,103], and mean μ and standard deviation sigma are 81.9 and 11.5 respectively.
1.84 σ are used as optimal intensity difference threshold value Tg.The reason for selecting multiplier 1.84 is the tonal gradation of all buildings all
In the range of 1.84 σ.
In present embodiment, reference data employs the building normal data (quilt of ISPRS Section III/4 working groups offer
Accurate classification is the experimental data of building point set and non-building point set), with the computational accuracy of quantitative assessment the method for the present invention.
The achievement of building object detecting method proposed by the present invention is represented in the form of volume elements, and the building in reference data
Object is expressed with discrete LIDAR laser point cloud datas, is and reference data compares to evaluate side proposed by the present invention
Method precision counts the number of the original airborne LIDAR cloud data included in the building volume elements that this method is detected, so first
It is compared afterwards with reference data and then with I classes error (building laser point mistake is divided into non-building laser point ratio), II
Class error (non-building laser point mistake is divided into building laser point ratio), the overall error (ratio of the building laser point of mistake point
Example), accuracy (the building laser points correctly detected account for the ratio of building laser point sum in testing result), integrity degree
(the building laser points correctly detected account for the ratio of building laser point sum in normal data), quality and Kappa coefficients
Carry out the validity of quantitative assessment building object detecting method proposed by the invention.
When table 1 is that spatial neighborhood scale is 6,8,26,56 and 80 in the present embodiment, the volume elements based on 2 experimental datas
Data are split and then are detected building therein, the precision index of corresponding building testing result.In the table
Data are intended to influence of the examination different field scale to building testing result, and thereby determine that optimal neighborhood scale.
The precision index of the building testing result of the different neighborhood scales of table 1
As shown in Table 1,6,18,26,56 and 80 average Kappa coefficients be respectively 34.5%, 66.5%, 71.3%,
89.1% and 87.3%, this explanation:The Kappa coefficients of (1) 56 neighbor assignment maximum, therefore, from the point of view of Kappa coefficient indexs,
56 neighborhoods are optimal neighborhood scales.(2) increase of neighborhood scale is not meant to the necessarily raising of accuracy of detection.The algorithm
Thought is that building information can be by being propagated based on the connectedness defined in volume elements array and intensity similarity.With 6 neighborhoods
Exemplified by, building information can only be propagated to 6 directions of its up, down, left, right, before and after, thus cause only flat-top (example
Such as, Area 2) on volume elements can just be integrated into a 3D connected region and be correctly detected, and positioned at fastigium buildings object (example
Such as, Area 3) on volume elements may be divided into multiple 3D connected regions and therefore by characteristics such as follow-up area, elevation saltus steps
And it judges by accident.This can explain why using 6 neighborhoods generate highest I classes error (the 4th different neighborhood scales of row in being shown in Table 1
Corresponding I classes error), it can also explain why the Kappa coefficients of the Area 2 of 6 neighborhoods are far above Area 3.With neighborhood
The increase of scale, the direction of propagation increase, and more volume elements are classified as building, are produced using 18,26,56 neighborhoods than 6 neighborhoods
Raw better result.But if neighborhood scale is too big, some non-building volume elements may be mistaken for building, therefore cause II
The increase (see the II types area of error 2 and 3 abutted using 80-) of class error, this can explain why 80 neighborhoods compare 56 neighborhoods
Precision declined instead.The average overall error of (3) 6,18,26,56 and 80 is respectively 25.8%, 14.9%, 12.8%,
5.1% and 5.9%, the average overall error of this explanation 56 neighbor assignments minimum.Therefore, from the point of view of overall error index, 56 neighborhoods
It is optimal neighborhood scale.
Step 4:Characteristic based on building roof and facade detects building roof and is connected with the 3D that facade is formed successively
Region completes the airborne LIDAR based on volume elements segmentation and builds analyte detection, and idiographic flow is as shown in Figure 9.
Step 4.1:Based on area performance, elevation hopping behavior and density feature, the 3D connected regions of building roof are detected
Building roof detection is completed in domain.
In present embodiment, the characteristic of building roof is:With certain area;There are elevations with surrounding landform
Difference;There are density variations in spatial distribution with other targets (such as vegetation).It is special according to more than area, elevation saltus step and density
Property establish the schemes of the corresponding 3D connected regions of search building roof from l 3D connected region:
Step 4.1.1:The 3D connected regions of non-building roof are rejected based on area performance.
In present embodiment, original airborne LIDAR cloud data is made to integrate the minimum floor area of building in P as Amin, make original
Airborne LIDAR cloud data integrates the floor area of building of the maximum in P as Amax。
The floor area of building in original airborne LIDAR cloud data collection P refers to projected area of the building in horizontal plane.
To any 3D connected regions, if its horizontal projected area is more than or equal to AminAnd less than or equal to Amax, then the 3D connect
Logical regional determination is building roof, is retained, otherwise rejects the 3D connected regions.
In the present embodiment, AminAnd AmaxFor constant, by user according to given original airborne LIDAR cloud data
Situation defines.
Step 4.1.2:The 3D connected regions of non-building roof are rejected based on elevation hopping behavior.
To any 3D connected regions, if the dispersed elevation of its contour line and the dispersed elevation of its surrounding terrain are more than what is given
Elevation threshold value Te, then the 3D connected regions be determined as building, retained, otherwise reject the 3D connected regions.Wherein, some
The dispersed elevation of the surrounding terrain of 3D connected regions can be obtained with following proposal:Utilize structural element [1 11;1 1 1;1 1
1] 3D morphological dilations are done to the 3D connected regions, the outer contour of note expansion process result is Cs={ va(ra, ca,
la), a=1 ..., q }, wherein, s-th of 3D connected region of behalf, a be s-th of 3D connected region outer contour in it is each
The index of volume elements, q are the volume elements numbers that the outer contour of s-th of 3D connected region includes.Search with
Its plane coordinates is identical and elevation is less than laNon-zero value volume elements, the dispersed elevation of above-mentioned volume elements is s-th of 3D connected region
The dispersed elevation of surrounding terrain;
In present embodiment, elevation threshold value TeFor constant, 2 meters of value.
Step 4.1.3:The 3D connected regions of non-building roof are rejected based on density feature.
To any 3D connected regions, if its density is more than given density threshold Td, then the 3D connected regions be judged to building
Object roof is built, is retained, otherwise rejects the 3D connected regions;
In present embodiment, density threshold TdFor constant, value can be according to the Density Distribution feelings of each 3D connected regions
Condition determines.
Step 4.2:Based on buffer zone analysis, the 3D connected regions of building facade are detected, complete the detection of building facade.
In present embodiment, the characteristic of building facade is:Perpendicular to building roof profile, positioned at building roof wheel
In wide surrounding's a certain range.According to These characteristics, determine to search the scheme of the corresponding 3D connected regions of building facade:
Step 4.2.1:Detect the profile of each building roof.
Step 4.2.2:In the horizontal plane, centered on any building roof profile, with each individual in inner side and outer side
Member establishes buffering area for width, as shown in Figure 10.
Step 4.2.3:To any 3D connected regions, if it is located inside buffering area and its gray value is corresponding with the region
The difference of the gray value of building roof is less than gray difference threshold Tg, then it is determined as building facade.
It is as shown in figure 11 using the building testing result obtained by the method for the present invention, wherein, (a) is the detection knot of Area2
Fruit, (b) are the testing result of Area3.Wherein, building testing result is buildings n-ary form n, and the size of single volume elements is
0.4m × 0.4m × 0.4m, as shown in the black cube in Figure 10.The building volume elements number point that Area2 and Area3 are detected
It Wei not be 25933 and 37589.Above-mentioned building volume elements is directly available to make building model, is a kind of 3D buildings of new model
Object meta-model.
Table 2 is in present embodiment, is standard to the building under 56 neighborhood scales of 2 experimental datas using reference data
The quantitative assessment that accuracy of detection carries out.
The precision of 2 building testing result of table
As shown in Table 2:Average quality, average integrity degree and the average accuracy for building analyte detection be respectively 89.2%,
89.7% and 96.8%.So as to demonstrate the validity of method proposed by the present invention.
Finally it should be noted that:The above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;To the greatest extent
Pipe is described in detail the present invention with reference to foregoing embodiments, it will be understood by those of ordinary skill in the art that:Its according to
Can so modify to the technical solution recorded in foregoing embodiments either to which part or all technical characteristic into
Row equivalent substitution;And these modifications or replacement, the essence of appropriate technical solution is not made to depart from the claims in the present invention and is limited
Fixed scope.
Claims (7)
1. a kind of airborne LIDAR based on volume elements segmentation builds object detecting method, which is characterized in that comprises the following steps:
Step 1:Original airborne LIDAR cloud data is read, forms original airborne LIDAR cloud data collection;
Step 2:Original airborne LIDAR cloud data rule is turned into gray scale 3D voxel data collection;
Step 2.1:The rejecting abnormalities data from original airborne LIDAR cloud data obtain removal abnormal data set;
Step 2.2:Removal abnormal data set rule is turned into gray scale 3D voxel data collection;
Step 3:Based on connective and radiation characteristic similarity criterion, gray scale 3D voxel datas are split and are labeled as several 3D
Connected region;
Step 4:Characteristic based on building roof and facade detects the 3D connected regions that building roof and facade are formed successively
Domain completes the airborne LIDAR based on volume elements segmentation and builds analyte detection;
Step 4.1:Based on area performance, elevation hopping behavior and density feature, the 3D connected regions of building roof are detected, it is complete
It is detected into building roof;
Step 4.2:Based on buffer zone analysis, the 3D connected regions of building facade are detected, complete the detection of building facade.
2. the airborne LIDAR according to claim 1 based on volume elements segmentation builds object detecting method, which is characterized in that institute
Step 2.1 is stated specifically to comprise the following steps:
Step 2.1.1:The frequency of each laser point height value in original airborne LIDAR cloud data is counted, and with histogram
Form visualization shows statistical result;
Step 2.1.2:Determine highest elevation threshold value T corresponding with real terrainhWith lowest elevation threshold value Tl;
Step 2.1.3:For each laser point in original airborne LIDAR cloud data, if its height value is higher than highest elevation threshold
Value ThOr less than lowest elevation threshold value Tl, then the laser point is abnormal data, is rejected, otherwise retains the laser point, finally obtain
Abnormal data set must be removed.
3. the airborne LIDAR according to claim 1 based on volume elements segmentation builds object detecting method, which is characterized in that institute
Step 2.2 is stated specifically to comprise the following steps:
Step 2.2.1:Three dimensions scope is represented with the oriented bounding box of removal abnormal data set;
Step 2.2.2:According to removal abnormal data the equalization point spacing of laser point is concentrated to determine point of volume elements in the x, y, z-directions
Resolution (Δ x, Δ y, Δ z), i.e. voxel size;
Step 2.2.3:Foundation voxel resolution (Δ x, Δ y, Δ z) divide oriented bounding box, obtain 3D volume elements grid,
Each 3D volume elements grid unit is volume elements;
Step 2.2.4:Each laser point is concentrated to be mapped to 3D volume elements grid removal abnormal data, and then according to 3D volume elements grid
In the Intensity attribute of laser point that includes be each volume elements assignment, obtain gray scale 3D voxel data collection.
4. the airborne LIDAR according to claim 1 based on volume elements segmentation builds object detecting method, which is characterized in that institute
Step 3 is stated to comprise the following steps:
Step 3.1:The non-zero value volume elements that gray scale 3D voxel datas are concentrated, the non-zero value body not being labeled until k-th are scanned successively
Member, wherein, k=1,2 ...;
Step 3.2:Using depth-first strategy traversal connected with k-th of non-zero value volume elements 3D and voxel values similar in all volume elements,
And labeled as Ll, wherein, l is the index of mark label, wherein, l=1,2 ...;
Step 3.3:Continue to scan on the not labeled non-zero value volume elements of gray scale 3D voxel datas concentration, until all volume elements all
It is labeled, obtain several 3D connected regions;
The difference of volume elements, that is, voxel values similar in the voxel values is less than two volume elements of gray difference threshold.
5. the airborne LIDAR according to claim 1 based on volume elements segmentation builds object detecting method, which is characterized in that institute
Step 4.1 is stated specifically to comprise the following steps:
Step 4.1.1:The 3D connected regions of non-building roof are rejected based on area performance;
Step 4.1.2:The 3D connected regions of non-building roof are rejected based on elevation hopping behavior;
Step 4.1.3:The 3D connected regions of non-building roof are rejected based on density feature.
6. the airborne LIDAR according to claim 6 based on volume elements segmentation builds object detecting method, which is characterized in that institute
Step 4.2 is stated specifically to comprise the following steps:
Step 4.2.1:Detect the profile of each building roof;
Step 4.2.2:In the horizontal plane, centered on any building roof profile, using each volume elements in inner side and outer side as
Width establishes buffering area;
Step 4.2.3:To any 3D connected regions, if it is located inside buffering area and its gray value building corresponding with the region
The difference of the gray value on object roof is less than gray difference threshold, then is determined as building facade.
7. the airborne LIDAR according to claim 3 based on volume elements segmentation builds object detecting method, which is characterized in that institute
It states as follows for the detailed process of each volume elements assignment according to the Intensity attribute of the laser point included in 3D volume elements grid:
Volume elements containing laser point is assigned a value of laser point strength mean value, the volume elements for not containing laser point is assigned a value of 0, further
By each volume elements assignment discretization to { 0 ..., 255 }, each voxel values are obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711362019.0A CN108074232B (en) | 2017-12-18 | 2017-12-18 | Voxel segmentation-based airborne LIDAR building detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711362019.0A CN108074232B (en) | 2017-12-18 | 2017-12-18 | Voxel segmentation-based airborne LIDAR building detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108074232A true CN108074232A (en) | 2018-05-25 |
CN108074232B CN108074232B (en) | 2021-08-24 |
Family
ID=62158483
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711362019.0A Active CN108074232B (en) | 2017-12-18 | 2017-12-18 | Voxel segmentation-based airborne LIDAR building detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108074232B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109711410A (en) * | 2018-11-20 | 2019-05-03 | 北方工业大学 | Three-dimensional object rapid segmentation and identification method, device and system |
CN111008989A (en) * | 2019-12-24 | 2020-04-14 | 辽宁工程技术大学 | Airborne multi-spectral LIDAR three-dimensional segmentation method based on multi-valued voxel |
CN112067633A (en) * | 2020-09-10 | 2020-12-11 | 张晓锋 | Steel construction building engineering quality intelligent detection system based on machine vision |
CN113689444A (en) * | 2021-07-07 | 2021-11-23 | 北京道达天际科技有限公司 | Building point cloud monomer segmentation method and device |
CN115079152A (en) * | 2021-03-10 | 2022-09-20 | 广达电脑股份有限公司 | Electronic device and method for positioning, imaging, detecting and distinguishing |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130155058A1 (en) * | 2011-12-14 | 2013-06-20 | The Board Of Trustees Of The University Of Illinois | Four-dimensional augmented reality models for interactive visualization and automated construction progress monitoring |
CN103322911A (en) * | 2013-05-27 | 2013-09-25 | 吴晓军 | Characteristic building elevation surveying and mapping method based on images |
US20150063683A1 (en) * | 2013-08-28 | 2015-03-05 | Autodesk, Inc. | Building datum extraction from laser scanning data |
CN105844602A (en) * | 2016-04-01 | 2016-08-10 | 辽宁工程技术大学 | Airborne LIDAR point cloud 3D filtering method based on volume elements |
US20160299997A1 (en) * | 2007-02-01 | 2016-10-13 | Autodesk, Inc. | Complete energy analytical model building information modeling (bim) integration |
CN106324581A (en) * | 2016-08-25 | 2017-01-11 | 辽宁工程技术大学 | Body-element-based airborne LIDAR building detection method |
CN106600710A (en) * | 2016-10-18 | 2017-04-26 | 浙江大学 | Layered machining multi-connected-domain construction method of adjacency topological correlation of geometric network model |
CN106709946A (en) * | 2016-12-16 | 2017-05-24 | 武汉大学 | Multiple-divided-conductor automatic extraction and fine modeling method based on LiDAR point clouds |
CN107169464A (en) * | 2017-05-25 | 2017-09-15 | 中国农业科学院农业资源与农业区划研究所 | A kind of Method for Road Boundary Detection based on laser point cloud |
-
2017
- 2017-12-18 CN CN201711362019.0A patent/CN108074232B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160299997A1 (en) * | 2007-02-01 | 2016-10-13 | Autodesk, Inc. | Complete energy analytical model building information modeling (bim) integration |
US20130155058A1 (en) * | 2011-12-14 | 2013-06-20 | The Board Of Trustees Of The University Of Illinois | Four-dimensional augmented reality models for interactive visualization and automated construction progress monitoring |
CN103322911A (en) * | 2013-05-27 | 2013-09-25 | 吴晓军 | Characteristic building elevation surveying and mapping method based on images |
US20150063683A1 (en) * | 2013-08-28 | 2015-03-05 | Autodesk, Inc. | Building datum extraction from laser scanning data |
CN105844602A (en) * | 2016-04-01 | 2016-08-10 | 辽宁工程技术大学 | Airborne LIDAR point cloud 3D filtering method based on volume elements |
CN106324581A (en) * | 2016-08-25 | 2017-01-11 | 辽宁工程技术大学 | Body-element-based airborne LIDAR building detection method |
CN106600710A (en) * | 2016-10-18 | 2017-04-26 | 浙江大学 | Layered machining multi-connected-domain construction method of adjacency topological correlation of geometric network model |
CN106709946A (en) * | 2016-12-16 | 2017-05-24 | 武汉大学 | Multiple-divided-conductor automatic extraction and fine modeling method based on LiDAR point clouds |
CN107169464A (en) * | 2017-05-25 | 2017-09-15 | 中国农业科学院农业资源与农业区划研究所 | A kind of Method for Road Boundary Detection based on laser point cloud |
Non-Patent Citations (3)
Title |
---|
曾齐红: "机载激光雷达点云数据处理与建筑物三维重建", 《中国博士学位论文全文数据库 信息科技辑》 * |
王丽英 等: "结合体元数据结构的机载 LIDAR 建筑物检测", 《中国图象图形学报》 * |
黄敏儿 等: "利用像素高度图的三维建筑物屋顶和立面提取方法", 《武汉大学学报 信息科学版》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109711410A (en) * | 2018-11-20 | 2019-05-03 | 北方工业大学 | Three-dimensional object rapid segmentation and identification method, device and system |
CN111008989A (en) * | 2019-12-24 | 2020-04-14 | 辽宁工程技术大学 | Airborne multi-spectral LIDAR three-dimensional segmentation method based on multi-valued voxel |
CN111008989B (en) * | 2019-12-24 | 2023-08-11 | 辽宁工程技术大学 | Airborne multispectral LIDAR three-dimensional segmentation method based on multivalued voxels |
CN112067633A (en) * | 2020-09-10 | 2020-12-11 | 张晓锋 | Steel construction building engineering quality intelligent detection system based on machine vision |
CN112067633B (en) * | 2020-09-10 | 2021-06-18 | 龙元明筑科技有限责任公司 | Steel construction building engineering quality intelligent detection system based on machine vision |
CN115079152A (en) * | 2021-03-10 | 2022-09-20 | 广达电脑股份有限公司 | Electronic device and method for positioning, imaging, detecting and distinguishing |
CN113689444A (en) * | 2021-07-07 | 2021-11-23 | 北京道达天际科技有限公司 | Building point cloud monomer segmentation method and device |
Also Published As
Publication number | Publication date |
---|---|
CN108074232B (en) | 2021-08-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111091105B (en) | Remote sensing image target detection method based on new frame regression loss function | |
CN110570428B (en) | Method and system for dividing building roof sheet from large-scale image dense matching point cloud | |
CN108074232A (en) | A kind of airborne LIDAR based on volume elements segmentation builds object detecting method | |
CN111145174B (en) | 3D target detection method for point cloud screening based on image semantic features | |
CN108109139A (en) | Airborne LIDAR three-dimensional building object detecting method based on gray scale volume element model | |
Wu et al. | Rapid localization and extraction of street light poles in mobile LiDAR point clouds: A supervoxel-based approach | |
CN112070769B (en) | Layered point cloud segmentation method based on DBSCAN | |
CN112099046B (en) | Airborne LIDAR three-dimensional plane detection method based on multi-value voxel model | |
CN106324581B (en) | A kind of airborne LIDAR building analyte detection method based on volume elements | |
CN105844602A (en) | Airborne LIDAR point cloud 3D filtering method based on volume elements | |
CN110047036B (en) | Polar grid-based ground laser scanning data building facade extraction method | |
CN109766824B (en) | Active and passive remote sensing data fusion classification method based on fuzzy evidence theory | |
CN113281782A (en) | Laser radar snow point filtering method based on unmanned vehicle | |
CN116452852A (en) | Automatic generation method of high-precision vector map | |
Boerner et al. | Voxel based segmentation of large airborne topobathymetric lidar data | |
CN117036971A (en) | Method for extracting airborne LiDAR data building under self-adaptive local spatial spectrum consistency | |
CN109785261B (en) | Airborne LIDAR three-dimensional filtering method based on gray voxel model | |
CN114004740B (en) | Building wall line extraction method based on unmanned aerial vehicle laser radar point cloud | |
Wolf et al. | Applicability of neural networks for image classification on object detection in mobile mapping 3d point clouds | |
Dey et al. | Building Boundary Extraction from LiDAR Point Cloud Data | |
CN111507423B (en) | Engineering quantity measuring method for cleaning transmission line channel | |
JP6018959B2 (en) | Data analysis apparatus, data analysis method, and program | |
Yastikli et al. | Automatic 3D building model generations with airborne LiDAR data | |
Huang et al. | Ground filtering algorithm for mobile LIDAR using order and neighborhood point information | |
Xiao | Detecting changes in trees using multi-temporal airborne LIDAR point clouds |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |